ray PROFESSOR TAIT ON THERMAL AND ELECTRIC CONDUCTIVITY. 
When the highest temperature (observed) was over 300° C., it was impossible 
to reconcile it with the curve deduced from the above formula from the indica- 
cations of the three succeeding thermometers. As this was obviously due to 
the rapid expansion of mercury near its boiling point, the irreconcilable obser- 
vation (sometimes as much as 10° above the curve mentioned) was not taken 
into account. 
§ 13. The Curves of Cooling were at first treated in the same way. But they 
had to be broken up into several sections, and it was not easy to decide 
(without great additional labour) how to obtain the most trustworthy value 
of the rate of cooling at a point common to two sections, from the more or less 
discordant values obtained from the separate formule for the sections. 
I next tried to treat them by taking three points with abscissz in arith- 
metical progression, and determining the common quantity to be subtracted 
from their ordinates, so that the intervening arc might be treated as loga- 
rithmic. |ForBEs used the logarithmic curve, but he made it pass through three 
points without subtraction from their ordinates. | 
This is a very good method so far as results go, and might be applied to all 
the different curves required for these experiments. But I found that, though 
the details which it involves are easy, even practised calculators were liable to 
get confused with their multiplicity. 
Finally, for my own revision of the whole work, I adopted the following 
method. I constructed a curve, usually with 5", 10™, and 20™ intervals for the 
abscissee, whose ordinates were 3th of the changes of temperature during the 5™ 
periods, or 5th of the changes for the 10" periods, &c. The scale for ordinates 
was usually much larger than that for abscissz. The points so determined did 
not, of course, give a very smooth curve (especially where successive readings 
at intervals of 1™ or 2™ came to be within one or two tenths of a degree of a 
division on the scale), but it was very easy to draw a smooth curve so as to 
equalize the errors, and the ordinates of this curve are at once the desired 
values of rates of cooling. This process has proved exceedingly successful. 
It is very much less tedious, and much less liable to large error, than any other 
at all accurate one—and its results compare favourably with those obtained 
by the other methods above. I believe that this process, applied to the cooling 
of bars, especially if one be of platinum, will give good results as to change of 
specific heat with temperature. 
I have already stated that as the short bars were always necessarily heated 
much above the temperatures at which their cooling was observed, my results 
are a little too large. The only really serious case is that of the copper bars. 
But for these the curve of cooling was observed through the same range for very 
different degrees of initial heating, and it was found that the only effect of 
oxidation was to increase all the ordinates through that range in a slowly 


